Gemini 3.5 Flash: Faster, Leaner, Built for Action
Sundar Pichai used the I/O stage to position Gemini 3.5 Flash as Google’s next big workhorse: a faster, more efficient AI model designed to blend “frontier intelligence with action.” Unlike heavyweight flagship models aimed at maximum reasoning depth, Flash is tuned for responsiveness and high-throughput use cases, where latency and cost matter as much as accuracy. It sits within the broader Gemini 3.5 family, but its purpose is clear: power real-time agents and interactive applications rather than just static chat. Tightly integrated with Google Antigravity, Google’s agent-first development platform, Gemini 3.5 Flash underpins a shift from AI that merely drafts content to AI that can carry out tasks across services. For developers and end users, that means more responsive assistants, smarter automation inside apps, and a foundation for large-scale multimodal AI search that feels instant instead of experimental.
Gemini Omni and the New Video Model: Multimodal AI Grows Up
Alongside Flash, Google introduced Gemini Omni, a model “that can create anything from any input, starting with video.” The new Gemini Omni video model is a major step in multimodal AI, extending understanding and generation from text and images into rich, dynamic video. Google describes Omni as a leap forward in world understanding, multimodality, and editing, signaling ambitions that go beyond simple clip tagging or transcript generation. In practice, this means more precise scene comprehension, timeline-aware reasoning, and the ability to edit or synthesize content using natural language instructions. Coupled with Gemini 3.5 Flash, Omni’s video capabilities form the backbone of new experiences like Ask YouTube, where users can query across long-form video directly. The strategy is to normalize video as a first-class input and output for AI, not an afterthought, bringing multimodal AI search closer to how people actually consume information.
Unified AI Overviews and AI Mode: Toward Seamless Multimodal Search
One subtle but important shift behind these Google I/O 2026 announcements is how search itself is being reimagined. Google is moving to unify AI Overviews and AI Mode into a single, seamless experience, reducing the friction between traditional search results and AI-generated answers. With Gemini 3.5 Flash handling fast reasoning and Gemini Omni powering richer multimodal understanding, the search box becomes less a link directory and more an interface to an always-on agent. Users can expect results that blend web pages, images, and now video understanding into one coherent answer, with the system able to act on their behalf where appropriate. This is multimodal AI search as infrastructure: instead of separate experimental features, AI Overviews unified with AI Mode will quietly underpin how people find, filter, and act on information across devices and contexts.
Agents Everywhere: Antigravity, Spark, and Universal Cart
The new Gemini models are tightly woven into Google’s broader push toward an “agentic” ecosystem. With advancements to Google Antigravity, its agent-first development platform, the company argues it has moved beyond AI tools that only help users write. Now, AI agents can help users act: Information agents in Search can proactively surface and organize knowledge; Gemini Spark and Daily Brief in the Gemini app aim to summarize and plan around ongoing tasks; and Universal Cart promises a truly intelligent shopping cart that can coordinate choices across merchants and sources. Gemini 3.5 Flash provides the speed and scalability for these interactions, while Gemini Omni contributes deeper multimodal understanding, particularly for video and complex media. Together, they mark a shift from isolated AI features to persistent, interconnected helpers that live inside search, productivity tools, commerce, and even intelligent eyewear.
Infrastructure and Strategy: Scaling Gemini With Cloud and TPUs
Underneath the product demos is a clear infrastructure play. Ahead of I/O, Google and Blackstone announced a major AI cloud venture meant to expand access to Google’s TPU-based compute. Backed by a USD 5 billion (approx. RM23.1 billion) initial equity investment from Blackstone, the new company will provide data centre capacity, operations, networking, and Google’s custom Tensor Processing Units as a compute-as-a-service offering. The first 500 megawatts of capacity are expected online in 2027, with plans to scale significantly over time. For Google, this serves two goals: meeting surging demand for AI computing power from partners like Anthropic and Meta, and ensuring it can support Gemini 3.5 Flash, Gemini Omni, and future models at global scale. The cloud venture underlines that Google’s AI strategy now spans everything from consumer-facing multimodal AI search to the industrial infrastructure that makes those capabilities possible.